Development of a novel method to measure bone marrow fat fraction in older women using high-resolution peripheral quantitative computed tomography.

Bone marrow adipose tissue (BMAT) has been implicated in a number of conditions associated with bone deterioration and osteoporosis. Several studies have found an inverse relationship between BMAT and bone mineral density (BMD), and higher levels of BMAT in those with prevalent fracture. Magnetic resonance imaging (MRI) is the gold standard for measuring BMAT, but its use is limited by high costs and low availability. We hypothesized that BMAT could also be accurately quantified using high-resolution peripheral quantitative computed tomography (HR-pQCT). METHODS: In the present study, a novel method to quantify the tibia bone marrow fat fraction, defined by MRI, using HR-pQCT was developed. In total, 38 postmenopausal women (mean [standard deviation] age 75.9 [3.1] years) were included and measured at the same site at the distal (n = 38) and ultradistal (n = 18) tibia using both MRI and HR-pQCT. To adjust for partial volume effects, the HR-pQCT images underwent 0 to 10 layers of voxel peeling to remove voxels adjacent to the bone. Linear regression equations were then tested for different degrees of voxel peeling, using the MRI-derived fat fractions as the dependent variable and the HR-pQCT-derived radiodensity as the independent variables. RESULTS: The most optimal HR-pQCT derived model, which applied a minimum of 4 layers of peeled voxel and with more than 1% remaining marrow volume, was able to explain 76% of the variation in the ultradistal tibia bone marrow fat fraction, measured with MRI (p < 0.001). CONCLUSION: The novel HR-pQCT method, developed to estimate BMAT, was able to explain a substantial part of the variation in the bone marrow fat fraction and can be used in future studies investigating the role of BMAT in osteoporosis and fracture prediction.

A deep multi-stream model for robust prediction of left ventricular ejection fraction in 2D echocardiography.

We propose a deep multi-stream model for left ventricular ejection fraction (LVEF) prediction in 2D echocardiographic (2DE) examinations. We use four standard 2DE views as model input, which are automatically selected from the full 2DE examination. The LVEF prediction model processes eight streams of data (images + optical flow) and consists of convolutional neural networks terminated with transformer layers. The model is made robust to missing, misclassified and duplicate views via pre-training, sampling strategies and parameter sharing. The model is trained and evaluated on an existing clinical dataset (12,648 unique examinations) with varying properties in terms of quality, examining physician, and ultrasound system. We report [Formula: see text] and mean absolute error = 4.0% points for the test set. When evaluated on two public benchmarks, the model performs on par or better than all previous attempts on fully automatic LVEF prediction. Code and trained models are available on a public project repository .

NoiseNet, a fully automatic noise assessment tool that can identify non-diagnostic CCTA examinations.

Image noise and vascular attenuation are important factors affecting image quality and diagnostic accuracy of coronary computed tomography angiography (CCTA). The aim of this study was to develop an algorithm that automatically performs noise and attenuation measurements in CCTA and to evaluate the ability of the algorithm to identify non-diagnostic examinations. The algorithm, "NoiseNet", was trained and tested on 244 CCTA studies from the Swedish CArdioPulmonary BioImage Study. The model is a 3D U-Net that automatically segments the aortic root and measures attenuation (Hounsfield Units, HU), noise (standard deviation of HU, HUsd) and signal-to-noise ratio (SNR, HU/HUsd) in the aortic lumen, close to the left coronary ostium. NoiseNet was then applied to 529 CCTA studies previously categorized into three subgroups: fully diagnostic, diagnostic with excluded parts and non-diagnostic. There was excellent correlation between NoiseNet and manual measurements of noise (r = 0.948; p < 0.001) and SNR (r = 0.948; <0.001). There was a significant difference in noise levels between the image quality subgroups: fully diagnostic 33.1 (29.8-37.9); diagnostic with excluded parts 36.1 (31.5-40.3) and non-diagnostic 42.1 (35.2-47.7; p < 0.001). Corresponding values for SNR were 16.1 (14.0-18.0); 14.0 (12.4-16.2) and 11.1 (9.6-14.0; p < 0.001). ROC analysis for prediction of a non-diagnostic study showed an AUC for noise of 0.73 (CI 0.64-0.83) and for SNR of 0.80 (CI 0.71-0.89). In conclusion, NoiseNet can perform noise and SNR measurements with high accuracy. Noise and SNR impact image quality and automatic measurements may be used to identify CCTA studies with low image quality.

Deep learning for [18F]fluorodeoxyglucose-PET-CT classification in patients with lymphoma: a dual-centre retrospective analysis.

BACKGROUND: The rising global cancer burden has led to an increasing demand for imaging tests such as [18F]fluorodeoxyglucose ([18F]FDG)-PET-CT. To aid imaging specialists in dealing with high scan volumes, we aimed to train a deep learning artificial intelligence algorithm to classify [18F]FDG-PET-CT scans of patients with lymphoma with or without hypermetabolic tumour sites. METHODS: In this retrospective analysis we collected 16 583 [18F]FDG-PET-CTs of 5072 patients with lymphoma who had undergone PET-CT before or after treatment at the Memorial Sloa Kettering Cancer Center, New York, NY, USA. Using maximum intensity projection (MIP), three dimensional (3D) PET, and 3D CT data, our ResNet34-based deep learning model (Lymphoma Artificial Reader System [LARS]) for [18F]FDG-PET-CT binary classification (Deauville 1-3 vs 4-5), was trained on 80% of the dataset, and tested on 20% of this dataset. For external testing, 1000 [18F]FDG-PET-CTs were obtained from a second centre (Medical University of Vienna, Vienna, Austria). Seven model variants were evaluated, including MIP-based LARS-avg (optimised for accuracy) and LARS-max (optimised for sensitivity), and 3D PET-CT-based LARS-ptct. Following expert curation, areas under the curve (AUCs), accuracies, sensitivities, and specificities were calculated. FINDINGS: In the internal test cohort (3325 PET-CTs, 1012 patients), LARS-avg achieved an AUC of 0·949 (95% CI 0·942-0·956), accuracy of 0·890 (0·879-0·901), sensitivity of 0·868 (0·851-0·885), and specificity of 0·913 (0·899-0·925); LARS-max achieved an AUC of 0·949 (0·942-0·956), accuracy of 0·868 (0·858-0·879), sensitivity of 0·909 (0·896-0·924), and specificity of 0·826 (0·808-0·843); and LARS-ptct achieved an AUC of 0·939 (0·930-0·948), accuracy of 0·875 (0·864-0·887), sensitivity of 0·836 (0·817-0·855), and specificity of 0·915 (0·901-0·927). In the external test cohort (1000 PET-CTs, 503 patients), LARS-avg achieved an AUC of 0·953 (0·938-0·966), accuracy of 0·907 (0·888-0·925), sensitivity of 0·874 (0·843-0·904), and specificity of 0·949 (0·921-0·960); LARS-max achieved an AUC of 0·952 (0·937-0·965), accuracy of 0·898 (0·878-0·916), sensitivity of 0·899 (0·871-0·926), and specificity of 0·897 (0·871-0·922); and LARS-ptct achieved an AUC of 0·932 (0·915-0·948), accuracy of 0·870 (0·850-0·891), sensitivity of 0·827 (0·793-0·863), and specificity of 0·913 (0·889-0·937). INTERPRETATION: Deep learning accurately distinguishes between [18F]FDG-PET-CT scans of lymphoma patients with and without hypermetabolic tumour sites. Deep learning might therefore be potentially useful to rule out the presence of metabolically active disease in such patients, or serve as a second reader or decision support tool. FUNDING: National Institutes of Health-National Cancer Institute Cancer Center Support Grant.

High-quality annotations for deep learning enabled plaque analysis in SCAPIS cardiac computed tomography angiography.

Background: Plaque analysis with coronary computed tomography angiography (CCTA) is a promising tool to identify high risk of future coronary events. The analysis process is time-consuming, and requires highly trained readers. Deep learning models have proved to excel at similar tasks, however, training these models requires large sets of expert-annotated training data. The aims of this study were to generate a large, high-quality annotated CCTA dataset derived from Swedish CArdioPulmonary BioImage Study (SCAPIS), report the reproducibility of the annotation core lab and describe the plaque characteristics and their association with established risk factors. Methods and results: The coronary artery tree was manually segmented using semi-automatic software by four primary and one senior secondary reader. A randomly selected sample of 469 subjects, all with coronary plaques and stratified for cardiovascular risk using the Systematic Coronary Risk Evaluation (SCORE), were analyzed. The reproducibility study (n = 78) showed an agreement for plaque detection of 0.91 (0.84-0.97). The mean percentage difference for plaque volumes was -0.6% the mean absolute percentage difference 19.4% (CV 13.7%, ICC 0.94). There was a positive correlation between SCORE and total plaque volume (rho = 0.30, p < 0.001) and total low attenuation plaque volume (rho = 0.29, p < 0.001). Conclusions: We have generated a CCTA dataset with high-quality plaque annotations showing good reproducibility and an expected correlation between plaque features and cardiovascular risk. The stratified data sampling has enriched high-risk plaques making the data well suited as training, validation and test data for a fully automatic analysis tool based on deep learning.

Semi-supervised learning with natural language processing for right ventricle classification in echocardiography-a scalable approach.

We created a deep learning model, trained on text classified by natural language processing (NLP), to assess right ventricular (RV) size and function from echocardiographic images. We included 12,684 examinations with corresponding written reports for text classification. After manual annotation of 1489 reports, we trained an NLP model to classify the remaining 10,651 reports. A view classifier was developed to select the 4-chamber or RV-focused view from an echocardiographic examination (n = 539). The final models were two image classification models trained on the predicted labels from the combined manual annotation and NLP models and the corresponding echocardiographic view to assess RV function (training set n = 11,008) and size (training set n = 9951. The text classifier identified impaired RV function with 99% sensitivity and 98% specificity and RV enlargement with 98% sensitivity and 98% specificity. The view classification model identified the 4-chamber view with 92% accuracy and the RV-focused view with 73% accuracy. The image classification models identified impaired RV function with 93% sensitivity and 72% specificity and an enlarged RV with 80% sensitivity and 85% specificity; agreement with the written reports was substantial (both κ = 0.65). Our findings show that models for automatic image assessment can be trained to classify RV size and function by using model-annotated data from written echocardiography reports. This pipeline for auto-annotation of the echocardiographic images, using a NLP model with medical reports as input, can be used to train an image-assessment model without manual annotation of images and enables fast and inexpensive expansion of the training dataset when needed.

Automatic pericardium segmentation and quantification of epicardial fat from computed tomography angiography.

Recent findings indicate a strong correlation between the risk of future heart disease and the volume of adipose tissue inside of the pericardium. So far, large-scale studies have been hindered by the fact that manual delineation of the pericardium is extremely time-consuming and that existing methods for automatic delineation lack accuracy. An efficient and fully automatic approach to pericardium segmentation and epicardial fat volume (EFV) estimation is presented, based on a variant of multi-atlas segmentation for spatial initialization and a random forest classifier for accurate pericardium detection. Experimental validation on a set of 30 manually delineated computer tomography angiography volumes shows a significant improvement on state-of-the-art in terms of EFV estimation [mean absolute EFV difference: 3.8 ml (4.7%), Pearson correlation: 0.99] with run times suitable for large-scale studies (52 s). Further, the results compare favorably with interobserver variability measured on 10 volumes.